HoVer

该数据集名为HOVER，包含了需要整合和推理多个维基百科文章的声明，并根据所需的“跳跃”次数被划分为不同的子集。验证集进一步根据推理深度（跳跃次数）被分为三个子集。具体来说，验证集中包含1126个需要两跳推理的声明，1835个需要三跳推理的声明，以及1039个需要四跳推理的声明。该数据集的任务是对复杂声明进行事实核查。

The dataset is named HOVER. It comprises claims that require integrating and reasoning across multiple Wikipedia articles, and is divided into different subsets based on the required number of "hops". The validation set is further subdivided into three subsets according to inference depth, which refers to the number of hops. Specifically, the validation set contains 1126 claims requiring two-hop reasoning, 1835 requiring three-hop reasoning, and 1039 requiring four-hop reasoning. The core task of this dataset is fact-checking complex claims.